Upsampling and the FR of a DAC


Hi everyone,
I've been playing around with Roon and currently have set it up to upsample (not oversample) 44.1 kHz to 88.2 and 48 to 96kHz. Kind of a nifty feature in that it lets you set up each source/target sample rate separately.

As you may know I've been happily living with a Mytek Brooklyn for a while, though it's not without issue.  I often run into problems with it not getting the correct sampling rate, or just not talking to the streamer via USB, but I digress.

As we are all clear, upsampling is the act of creating a new, higher sample rate signal from the original, but we cannot create more information, we are just smoothing it out. That is, any overtones which were not present in the original 44.1 kHz signal won't be magically recreated in the upsample. Maybe someday AI will do this, but for now this is how upsampling for audio signals work.
HOWEVER!!! I have a question I hope the more technically inclined can help answer.

Looking at the frequency response (FR) plot of the Stereophile measurements here:
https://www.stereophile.com/content/mytek-hifi-brooklyn-da-processor%C2%96headphone-amplifier-measur...

It seems the response at 20 kHz is significantly different for a 44.1 kHz signal vs. 96 kHz (and I think we can assume 88.2 would be in between).  Lets use the second chart, the slow rolloff filter as an example.  At 20 kHz 44.1 is down about 2.25 dB but at 96 kHz only about a tenth of that.

So here is my question.  If we upsample from Redbook (44.1/16) do we still get the 96 kHz frequency plot as a result?  If so, couldn't this alone be a possible reason why some hear a difference in upsampling and some do not?
erik_squires
If we upsample from Redbook (44.1/16) do we still get the 96 kHz frequency plot as a result? If so, couldn't this alone be a possible reason why some hear a difference in upsampling and some do not?
Yes, and yes. But are SR (slow rolloff) filters very popular? The vast majority of linear phase filters won't have such a droop approaching 20kHz as the SR one here.
Yes, and yes. But are SR (slow rolloff) filters very popular?


I don't use it, I was just using the graph as a convenient example of the difference in the frequency response at different data rates.  The point I wanted to ask was more difficult to make with the fast rolloff filter chart.

My understanding is if you do upsample, then stay within the same "family". For example, 44.1, 88.2, 176.4, 352.8 or 48, 96, 192, 384.

Yep, this makes the math simpler, and you retain the original data exactly, though as Benchmark has written, and perhaps discovered, if you attempt to use a curve fitting algorithm adding bits for overhead is necessary.

Otherwise,  you end up in asynchronous sample rate conversion, which some DACs now do to solve the jitter problem altogether, while playing a little loose with the data.

Best,
E